[arnaud-lb]

Broke the build while merging #24585 up, as I didn't anticipate that the header needs to be C++ compatible when building intl.

I'm wondering if we should eventually re-introduce a smaller zend_atomics.h header for C++ compat.

For now I would like to fix the build.

cc @henderkes @devnexen

[devnexen]

Suggested change

	#ifdef __cplusplus
	#ifdef __cplusplus
	static_assert(sizeof(std::atomic<uint32_t>) == sizeof(uint32_t), "atomic size");
	static_assert(alignof(std::atomic<uint32_t>) == alignof(uint32_t), "atomic align");
	static_assert(ATOMIC_INT_LOCK_FREE == 2, "not always lock-free");
	uint32_t type_info = reinterpret_cast<const std::atomic<uint32_t> *>(&GC_TYPE_INFO(s))->load(std::memory_order_acquire);
	#else
	static_assert(sizeof(_Atomic uint32_t) == sizeof(uint32_t), "atomic size");
	static_assert(alignof(_Atomic uint32_t) == alignof(uint32_t), "atomic align");
	static_assert(ATOMIC_INT_LOCK_FREE == 2, "not always lock-free");
	uint32_t type_info = atomic_load_explicit((_Atomic uint32_t *)&GC_TYPE_INFO(s), memory_order_acquire);
	#endif

[arnaud-lb]

Thank you. I've pushed a simplified version before seeing this comment because MSVC also doesn't have ATOMIC_INT_LOCK_FREE == 2 in C.

[devnexen]

looks ok per see.

[henderkes]

I'm wondering if we should eventually re-introduce a smaller zend_atomics.h header for C++ compat.

I think a file just for atomics because of C++ is overkill, perhaps a typedef (#ifdef cplusplus) in zend_portability.h would do?

[arnaud-lb]

@henderkes that would be ideal indeed

[arnaud-lb]

I'm going to merge this to fix the build for now

[LamentXU123]

@arnaud-lb I think you should also backport this to 8.6?

[arnaud-lb]

@LamentXU123 is 8.6 broken too?

[LamentXU123]

Oh my bad. I look into the wrong branch. Sry :)